Professor Steven Verstockt and his research team from Ghent University are working together with the International Cycling Union (UCI) and the new safety organization SafeR to map falls in races and screen courses. “Flanders Classics is among the best students in the class when it comes to safety,” says Verstockt.
What is the role of your research team within the safety issue?
Steven Verstockt: At the end of 2020, we entered into a partnership with the International Cycling Union UCI. We were tasked with creating a database of all accidents that had occurred in the most important WorldTour races (the three Grand Tours and the five classic monuments) and the World Cup since 2016. We use the latest technologies for this purpose machine learning and natural language processing in. We use a scraper, a software technique, to search for messages about price incidents on X (the former Twitter, ed.). These tweets are then linked via an algorithm to a race, which riders are involved, where the incident occurred… We have developed a fully automated database, which we update every week.
In addition, we receive from the UCI, and since last year also through the new one safety organization SafeR, each season a list of competitions where we screen the course a few weeks in advance. An algorithm maps the dangerous, high-risk zones based on GPX data and video images and our incident database. We will pass this on to the UCI and the organizers. It is up to them to take any additional safety measures.
Shouldn’t this course screening be done at least at every WorldTour race?
Verstockt: I don’t know exactly what criteria the UCI uses for the list of races that we have to screen. The intention is for that list to be expanded next season. There are still many practical problems with supplying video images of the course. A lot of time is also lost setting up meetings with the various stakeholders to discuss the results.
That is why we are now considering an automated software package to better streamline the entire process. This allows organizers to detect their risk zones themselves, with their own dashcam, and intervene before they submit their course to the UCI.
What innovations are you still working on?
Verstockt: Of large language models, everything related to ChatGPT, we can search even more specifically on different platforms for causes and the impact of incidents in the days after a match. Now that only happens during a race.
Furthermore, in addition to the course screening a few weeks before the race, live screening could also be very useful. Then you ride the course an hour before the riders and monitor with a live dashboard where there is oil or stones on the road, which roads are dangerously slippery due to bad weather, and which slope or place on a col there are too many people. so that you can close off certain zones if necessary. We have already experimented with that. And that can also be introduced at relatively short notice. At least in terms of technology, because organizers must also employ people who can complete the course. If they can link a commercial model to this, they may be more inclined to do so.
Also read: SafeR project starts this season: cyclists pay for their safety
Is it difficult to convince organizers?
Verstockt: The technology, especially now with the implementation of artificial intelligence, is available to collect even more information and thus increase safety. However, organizers sometimes still lack the will and resources to make use of it. It is the biggest challenge of the security problem. Foreign organizations in particular, including major World Tour races, still have too much their own vision and opinion, the belief that they themselves have sufficient knowledge. Given all the developments in cycling, you can no longer judge courses as you did ten or twenty years ago. A big step forward has already been taken with SafeR. Hopefully all organizers will be convinced that we should embrace this new technology.
By the way, that is the sad thing about the crash in Dwars door Vlaanderen. Organizer Flanders Classics is among the best students in the class in terms of safety, in many areas. The criticism of them is not entirely justified. They cannot prevent everything either. And neither did we, because Dwars door Vlaanderen or the Tour of Flanders were not on the UCI screening list. Flanders Classics has decided on its own to cancel the Kanarieberg in the Ronde. A good decision, it now appears.
😱 A huge crash in the leading group has taken down a number of favorites, including Wout van Aert, Mads Pedersen and Biniam Girmay, at Dwars door Vlaanderen 💥🤕
Sending our best wishes to the riders who were involved in the crash. pic.twitter.com/hlfuZwhBZd
— Eurosport (@eurosport) March 27, 2024
Should organizers be required to remove sections that you label as ‘high risk’ from their course?
Verstockt: At the very least, organizers should think more about alternatives to dangerous roads. However, this is not possible on every course. For example, the descent of the Cipressa and the flat section towards the Poggio cannot be achieved from Milan-San Remo, but there are more alternative options in the Flemish Ardennes. If this is not possible, organizers must identify dangerous points as clearly as possible, before and during the competitions.
Still, you can’t put all the blame on the organizers. Team leaders and riders also have to look into their own hearts. Ultimately, they determine how much risk is taken. Even the media should point out more dangerous maneuvers in their reporting. When Matej Mohoric won Milan-San Remo in 2022, his descending abilities and his innovative dropper post were loudly applauded, while he was very lucky twice and could just as easily have fallen.
Racing has become increasingly offensive in recent years. Can you measure to what extent this contributes to falls?
Verstockt: Monitoring race behavior and the dynamics of a peloton is currently not possible. Well, if we could measure the behavior of each rider with a sensor/accelerometer, under the saddle or pinned on a bib number. We can now detect falls through our current tools, but not near crashes or when riders suddenly brake hard. This data could also be useful for identifying dangerous situations and points. They are now available via the riders’ cycling computers, but it is cumbersome to collect that data from the teams, if they even want to do so. With a uniform sensor the procedure would run much smoother.
Two years ago you announced the results of your research into falls. Have there been any noticeable developments since then, in terms of numbers, times of matches and/or places?
Verstockt: No, you can still draw the same conclusions. With the nuance that Elon Musk has not made it easy for us to collect information via X, which means we have less data for a certain period. Nevertheless, the data shows that cases are not increasing, but not decreasing either.
In terms of times and places, you have also seen the same logical observations recurring in recent years: almost half of the crashes take place in the final, in the last twenty kilometers, and another third between kilometers sixty and twenty.
The reasons for a fall also remain the same: descents are number one, the run-up to one point of interest – a cobblestone section, slope, intermediate sprint… – on two, and mistakes by riders on three, for example rear-end collisions. It is no coincidence that this mix caused the crash towards the Kanarieberg on Wednesday in Dwars door Vlaanderen.
